View
1
Download
0
Category
Preview:
Citation preview
1
Working papers in Economics No 169
Contingent Valuation of Mortality Risk Reduction in
Developing Countries: A Mission Impossible? �
Minhaj Mahmud* Department of Economics, Göteborg University
Vasagatan 1, Box 640, SE- 405 30 Göteborg, Sweden
April 2005
�$EVWUDFW�Using the contingent valuation method in developing countries to value mortality risk
reduction is particularly challenging because of the low level of education of the
respondents. In this paper, we examine the effect of training the respondents regarding
probabilities and risk reductions, in addition to using visual aids to communicate risk
and risk reductions, in a contingent valuation survey. Our results indicate a significantly
higher WTP for the trained sub-sample, and WTP is sensitive to the magnitude of risk
reduction both with and without the training.
.H\ZRUGV: contingent valuation; risk reduction; WTP; effect of training; sensitivity to
scope; Bangladesh.
-(/�&ODVVLILFDWLRQ: I1, D6, D8, H4
* I am most grateful for very useful suggestions and comments from Olof Johansson-Stenman, Peter
Martinsson, Fredrik Carlsson, Håkan Holm, and seminar participants at Göteborg University. Financial
assistance from the Swedish International Development Agency (SIDA) is also gratefully acknowledged.
2
���,QWURGXFWLRQ�The contingent valuation method (CV) has been widely used to value mortality risk
reduction, but mostly in developed countries (e.g. Corso et al., 2001; Persson et al.,
2001; Krupnick et al., 2002). The CV method involves eliciting people’s willingness-to-
pay (WTP) for a hypothetical reduction in the risk of dying during a given time period
(see Hammitt and Graham, 1999). The individual’s rate of trade-off between own
money and a small risk change1 i.e. the marginal rate of substitution, is defined as the
value of a statistical life (VSL) (Weinstein et al., 1980; Viscusi, 1993; Johansson, 2002).
The individuals’ WTPs for a reduction in the mortality risk are converted to VSL by
dividing the WTPs by the risk change in question. However, most previous CV studies
have found unreasonably low sensitivity of WTP to the size of the risk reduction. One
likely reason for the lack of sensitivity is a poor understanding of probabilities and a
lack of intuition regarding small risk changes (see Hammitt and Graham, 1999).
However, recent evidence by Corso et al. (2001) suggests that there are ways to increase
the sensitivity by using visual aids in the presentation of risks in the CV survey.
It is particularly challenging to use the CV method to value mortality risk
reductions in developing countries.2 The main reason is the difficulty to communicate
probabilities and risk reduction to the respondents, since many either have very low
levels of education or are illiterate. Moreover, most people are unfamiliar with the
concept of trading income for the risk reduction and therefore might face greater
uncertainty in placing a value on the risk reduction. A brief training of the respondents
in the survey regarding probability, risk, and risk changes may enable the respondents to
1 For example, a change in risk of dying from 5 in 10,000 to 4 in 10,000. 2 However, by a careful survey design, it is generally possible to conduct high quality CV surveys in
developing countries (see Whittington, 1998; 2002).
3
process risk information better and thus the respondents will become more elaborate
about their preferences for risk reduction. Therefore, one might expect that by reducing
any uncertainty regarding the object of valuation, the training would yield lower
variation in the responses, as well as an increase in the sensitivity to scope. This paper
reports on a CV study of mortality risk reduction conducted among a random sample of
rural households in Bangladesh. The objective of the study was to examine the effect of
training regarding probability, risk, and the implication of risk reductions on the WTP
responses and not to obtain an absolute magnitude of the VSL, as well as to investigate
whether training affects the sensitivity to scope.
The validity and reliability of the CV method is intensely debated (see e.g.
Kahneman and Knetsch, 1992; Diamond and Hausman, 1994; Hanemann, 1994).
However, some of the criticisms attached to the CV method, such as “warm glow,” or
the “purchase of moral satisfaction” phenomenon for contributing to overall social
causes, is not applicable when valuing individuals’ own risk reductions through
vaccinations.3 On the other hand, it can be highly cognitively demanding for the
respondents to compare expected welfare effects from the risk reductions to the effects
of monetary changes (Beattie et al., 1998; Hammit and Graham, 1999; Carlsson et al.,
2004). The sensitivity of the estimated WTP to the magnitude of the good in question is
regarded as a test of the validity of CV estimates (Mitchell and Carson, 1989; NOAA,
1994; Diamond and Hausman, 1994). Assuming that risk reduction is a desired good,
the theoretical expectation is that WTP for mortality risk reduction should be positively
associated with the magnitude of the risk reduction. Furthermore, for sufficiently small
risk changes, WTP should be proportional to the magnitude of risk reduction (Weinstein 3 Although vaccinations can be seen as a good with a positive externality, we find no indication that
people consider this while deciding for their own vaccination.
4
et al., 1980; Hammitt and Graham, 1999; Hammitt, 2000). This means e.g. that WTP
should be twice as high for a two-fold reduction in risk. However, a problem of stating
risk in a survey is when the respondents treat the given probabilities as not applicable to
them and hence form posterior risk estimates based on their prior beliefs and on
information contained in the scenario. In this case, the stated WTP will not be
proportional to the magnitude of the stated risk reduction presented; rather, it would
rather be proportional to changes in the perceived risk (Viscusi, 1985; 1989).
The use of visual aids has proven to be useful (e.g. Corso et al., 2001; Krupnick
et al., 2002) in obtaining responses consistent with the theoretical expectations
(sensitive to scope). Examples of visual aids include verbal probability analogies, risk
ladders, and graph paper with squares and an array of dots. Corso et al. (2001) found
that graph paper and a risk ladder with logarithmic representation of the risk performed
the best.
Accordingly, we used graph paper to communicate risk and risk reduction in our
survey. We stated either a 25% or a 50% reduction in the risk that corresponds to the
respondents’ subjective risk of dying during the next five years; subjective risk is based
on age-related statistical risks of dying for the same period presented in the CV
scenario. Therefore, the insensitivity of scope problem can be expected to be smaller in
view of the fact that we include a relatively large risk change together with training on
risk reduction, particularly if the insensitivity is related to the poor understanding or
lack of intuition about small risk changes on the part of the respondents. Another reason
why the theoretical proportionality prediction would not hold in our case is that we deal
with substantial risk changes; WTP would increase but less than proportionally to the
risk changes and hence the resulting VSL is expected to be smaller compared to the case
5
when small risk changes are valued. The results indicate a significant difference in the
distribution of the WTP between the sub-sample receiving training and the sub-sample
receiving no training. However, we find no significant difference in the variance, but
rather that WTP increases with training. We also find that estimated WTP is sensitive to
the scope of the risk reduction, both with and without training.
The remainder of the paper is organized as follows: Section 2 presents the
design of the CV survey including sample characteristics, respondents’ risk perceptions
etc., Section 3 discusses WTP results, Section 4 presents the econometric analysis, and
the paper is concluded in Section 5.
�
���'HVLJQ�RI�WKH�&9�VXUYH\��Two versions of the CV survey were constructed: one version including a brief training
vis-à-vis probability and risk, and the other without such training.4 The enumerators,
used to conduct the survey, were trained beforehand using the guidelines of Whittington
(2002). In particular, the enumerators were trained regarding the risk presentations, and
the CV methodology in brief, i.e. the purpose of the CV survey, the notion of maximum
WTP, etc. The same enumerators were used in the pilot and in the final survey, and they
were closely supervised during the fieldwork. We furthermore test for possible
enumerator effects when analyzing WTP responses. The CV questionnaire was tested
using focus groups and two pilot studies, which, together with the feedback from the
enumerators, enabled us to simplify the risk presentations, the CV scenario and the CV
question. The survey-questionnaire and the CV scenario were translated back to English
4 Ideally, it would be better to train people for a longer period; however, we intend to see whether a brief
training as part of the questionnaire makes any significant difference in the responses since such
education is the most realistic kind that can be pursued.
6
from Bengali to ensure the exact meaning of the original English version. The final
survey was administered, using a random sample technique, among rural households in
the following five districts of Bangladesh: Netrokona, Mymensingh, Manikganj,
Gazipur, and Narayanganj. The sample is therefore not representative of the Bangladesh
population, which consists of 64 districts. Moreover, the villages were chosen so that
the respondents of the Hindu religion are over-represented (compared to the national
average of 11 %), in order to facilitate religious comparisons. The enumerators were
allocated to different parts of the selected villages and were then asked to perform
household surveys and the CV experiments. The interviews were conducted with the
household head as the decisions made within households are normally made, or at least
approved of, by the household head. If a household head was not around, the
enumerators were instructed to return later. If the respondent was not home at the
second visit, the enumerator moved to next household.5 The participation rate of
household heads approached for interviews was 99 %. The respondents were paid (100
Taka) as an appreciation of their time and cooperation. Table 1 presents the sample
statistics for the full sample. The household survey also included detailed questions on
respondent health and risk perceptions, in addition to the socioeconomic questions. The
CV survey took on average 15 to 30 minutes to complete. 774 individuals were
interviewed. Table 1 presents the sample statistics.6
5 However, in the villages people from the same family-chain normally live in a cluster of 4-5 households.
Thus, a replacement from the next household or next to the next household (in some cases) should not
bias the results. There were 22 % replacement households in our sample. 6 We have excluded observations related to very old individuals (older than 75), as the risk reduction
presented in the CV survey is for a five -year period.
7
>>> TABLE 1 HERE
The respondents were asked to state their maximum WTPs for obtaining a stated
risk reduction that corresponded to their stated subjective risk of dying during the next
five years, which was elicited after the average age-related objective risk had been
presented to them. We choose the open-ended format as it provides more information
than the closed-ended (dichotomous choice question) format, although many researchers
would favor the latter (see e.g. Bateman et al., 2002; Hanley et al., 2003). Moreover, it
has been shown in experiments, that dichotomous choice overestimates values more
than the open-ended questions in the case of auction values as well as private goods (see
Balistreri et al., 2001).
Based on a W� test, we do not find any statistically significant differences (p-
value>0.05) in terms of socio-economic characteristics between the populations of the
two sub-samples, i.e. training and no training. Therefore, differences in the WTP
responses (relating to a specific risk reduction), between these two sub-samples could
be attributed to receiving training in the survey.
The responses can be divided into the following categories: (1) training and a
50% risk reduction, (2) training and a 25% risk reduction, (3) no training and a 50% risk
reduction, and (4) no training and a 25% risk reduction.
����7UDLQLQJ�DQG�ULVN�XQGHUVWDQGLQJ�In the CV questionnaire, the training involved concepts of probability of different
events occurring, risks, and implications of risk changes (presented in Figure 1). In
particular, we used coin flipping, dice throwing and a lottery example to introduce the
8
concept of probabilities to the respondents. Mortality risk was discussed using the
example of risk of dying from traffic accidents. The chance of winning in a lottery and
the mortality risk example were explained with the use of graph paper containing 100
and 1,000 squares, respectively. The respondents were asked test questions after each
example. If the respondent had a correct answer, the enumerator continued to the next
example. To facilitate understanding the respondents received more explanation
following a wrong answer, before being asked the same question again. If a respondent
still did not have a correct answer after the third, then the enumerator continued to the
next example after explaining the correct answer.
>>> FIGURE 1 HERE
The results of the probability test questions are summarized in Table 2. A respondent is
considered to have passed the entire test if he/she provided the correct answer to each of
the three probability test questions on the first attempt. Only 24% of the respondents
passed the entire test.
>>> TABLE 2 HERE
�
In the second stage of the training, the meaning of the risk reductions was
explained to the respondents (presented in Figure 2, read by the enumerators). To begin
with, the respondents were informed about the average risk of dying for an adult in
Bangladesh in the next five years (40 in 1,000).7 This risk was explained using a graph
paper containing 1,000 squares of which 40 were colored black (see Appendix). Then 7 This is based on Bangladesh life table estimates for the year 2000 provided by the World Health
Organization (WHO, 2004).
9
the respondents were told that with appropriate public policy this mortality risk could be
reduced to for example 35 in 1,000. The two risk levels were shown simultaneously
using graph papers to explain the differences with five of the black squares becoming
white in the second graph.
>>> FIGURE 2 HERE
The implication of the risk changes from 40 in 1,000 to 35 in 1,000 was
explained to the respondents by saying that 5 out of 40 lives could be saved through a
policy measure. In a similar fashion, the meaning of further risk reduction was
explained to the respondents, i.e. reducing the risk from 40 in 1,000 to 20 in 1,000 and
reducing the risk from 40 in 1,000 to 10 in 1,000. Each risk reduction example was
explained up to three times to facilitate the respondents’ understanding. Almost 95% of
the respondents revealed that they had understood all risk reduction examples after the
first explanation. This might reflect a “ yea saying” bias as the question (Do you
understand this risk reduction?) is of a yes/no type. However, when asked in the end of
the training to indicate which of the above three risk reduction examples they would
prefer, almost 98 % of the respondents preferred the largest risk reduction example
(reducing the risk from 40 in 1,000 to 10 in 1,000), which suggests that they had
understood the risk reduction examples.
�
����2EMHFWLYH�ULVN�DQG�ULVN�SHUFHSWLRQ�All respondents in the survey, before being presented with the CV scenario, were first
informed about the average mortality risk of persons aged 30-34 and persons aged 55-
59 as 15 in 1,000 and 90 in 1,000, respectively, in the next five year period (see Figure
10
3). Then the respondents were asked to mention their perceptions of their own risks of
dying during the same period taking into consideration their ages, health and lifestyles
in particular.
>>> FIGURE 3 HERE�
Thus, we customize the mortality risk for each individual according to his or her
own perception. Figure 4 shows the mean mortality risk, both objective and subjective,
for various age groups of respondents during the next five years. As observed, people on
average overestimate mortality risk at younger ages and underestimate it at older ages.
This supports earlier findings (Viscusi, 1992) that people tend to overestimate small
risks and underestimate large risks. Based on a non-parametric (Wilcoxon matched-
pairs signed-rank) test, we can conclude that respondents’ subjective and objective (age-
related) risks are significantly different (p- value <0.001).8 That people have a biased
risk perception is also consistent with much research in psychology (Tversky and
Kahneman, 1992; Kahneman and Tversky, 2000; Gilovich et al., 2002).
>>> FIGURE 4 HERE
We estimate ordinary regression to see what characteristics explain the risk
perception of individuals. For obvious reasons we do not focus on gender differences in
risk perception since we only have 9% female respondents.9 As age and age-related
objective risk are highly correlated, we estimate two separate models. The first model
8 The difference between objective and subjective risk is positive, negative and zero in 446, 290 and 35
cases, respectively. 9 We do not focus on gender difference in any of the subsequent analyses of this paper.
11
includes age-related objective mortality risk and the second model includes
respondents’ ages, in addition to other explanatory variables. Based on a PE test
(Greene, 2000), we can reject the null hypothesis of a linear specification in favor of a
log-linear specification (PE coefficient 70.61, p-value 0.000), in the case of the first
model. For the second model, however, we cannot reject neither the null hypothesis of a
linear specification (PE coefficient 18.97, p-value 0.285), nor the null hypothesis of a
log-linear specification (PE coefficient 0.010, p-value 0.335). In Table 3, we present the
results of log-linear specifications for both models.
>>> TABLE 3 HERE
We observe that the respondents’ risk perception increases by only 0.5 % for a
1% increase in the average age-related objective risk. We find that respondent’ s health
status significantly affects the perception of the risk of dying; people with chronic
illness show a 16% higher risk perception (first model) compared to people not having
any chronic illness. Although weakly significant, the Muslim respondents seem to have
a 15% higher risk perception compared to the Hindu respondents.
We further observe that the smokers’ perception of the risk of dying is 19% higher
compared to non- smokers. Thus, it seems like smokers are quite aware of the health
risk of smoking even in the rural areas of a developing country such as Bangladesh. On
average, smokers’ life expectancy is shorter than for non-smokers. For example, Shaw
et al. (2000) estimate that average loss of life due to smoking is 6.5 years or 11 minutes
per cigarette. Studies have shown that the risk of developing lung cancer is 22% higher
for smokers and that the mortality risk from cardiovascular disease (heart disease) is
almost double for smokers (including ex-smokers) compared to non-smokers
12
(Newcomb and Carbone, 1992; ILO, 2002).10 Given these estimates, it is hard to say
whether smokers overestimate or underestimate the risk of dying in our case. The
empirical evidence regarding smokers’ risk perception is otherwise mixed.11
����7KH�&9�VFHQDULR��Finally, the respondents were presented with the CV scenario (see Figure 5) and WTP
questions to elicit their preferences for a risk reduction. The specific risk reduction
(either a 25% or a 50% reduction) corresponds to the respondent’ s stated perceived risk
of dying during the next five years and was communicated by separating the black
squares representing perceived risk into 25-75 or 50-50 splits. The risk reductions, as
told to the respondents, were to be achieved through participation in a program
involving various vaccinations. Each respondent was asked about his/her maximum
WTP for the stated risk reduction. If the respondent stated zero WTP for the risk
reduction, he or she was asked several follow-up questions in order to ascertain possible
scenario rejections.
>>> FIGURE 5 HERE
� 10 The risk for contracting other types of cancer is also relatively higher for smokers (Newcomb and
Carbone, 1992). 11 For example, studies in developed countries find that smokers overestimate the risk of getting lung
cancer from smoking and that their assessed loss in life expectancy due to smoking is quite high (Viscusi,
1990; Viscusi 1992). For a sample of smokers in Sweden, Hammar and Johansson-Stenman (2004),
however, did not find support for the conclusion that smokers overestimate the health risk from smoking.
Slovic (2000) discusses the fact that particularly young smokers considerably underestimate the health
risk due to smoking.
13
����:73�UHVXOWV�Approximately 10 % of the respondents stated zero WTP. Those who stated zero WTP
were asked if they instead would want to receive the vaccination free of cost; 77 % of
respondents with zero stated WTP mentioned that they would rather want the risk
reduction free of cost. To ascertain scenario rejection, all these respondents (including
the ones who would not want free vaccination) were asked why they would not be
willing to pay for vaccinations; possible answers (reasons) were read to the respondents
and the respondents could choose more than one answer (see Table 4). Among those 79
respondents, 84 % had chosen more than one answer; we believe these responses
indicate scenario rejection.12 We do not include these responses in our further statistical
analysis of this paper.
>>> TABLE 4 HERE
However, we analyze the probability of scenario rejection using a standard
probit model. In Table 5, we see that the only significant variable that explains scenario
rejection is the respondent’ s religious belief; Muslim respondents are more likely to
provide a protest zero. We also see that respondents receiving training in the survey are
less likely to reject the scenario at the 10% level.
>>> TABLE 5 HERE
12 Respondents who had chosen any response than (i) or had chosen more than one responses are believed to have provided protest zeros when answering the WTP question.
14
Alberini (2004) discusses about the robustness of CV estimates and different
types of outliers. We look for WTP outliers in relation to income. There is no D�SULRUL reason to assume that WTP for reducing the risk of dying should be a small part of
income, since the payment for the risk reduction was to be made once for a five-year
period. We choose to keep WTP responses equal to or less than 50% of respondent
annual income per capita. The coefficient of income remains roughly the same with this
exclusion.13 We finally have 692 observations for further statistical analysis. Table 6
reports the descriptive statistics of WTP for different sub-samples; in the Appendix, we
present histograms of the distribution of WTP, where a visual inspection of WTP data
reveals the existence of some focal points.
>>> TABLE 6 HERE
We find that the mean WTP is TK. 487 (TK. 672) for a 25% (50%) risk
reduction, for the no-training sub-sample. For the trained sub-sample, the mean WTP is
TK. 671 (TK. 970) for a 25 % (50%) risk reduction. Thus, we find that the training in
the CV survey increases the mean and reduces the variances of WTP. We can reject the
hypothesis that for the specific risk reduction (either 25% or 50%), WTP for the two
sub-samples (training and no training) comes from the same underlying distribution (in
both cases, p-value<0.001). As reported in Table 6, the results also indicate a smaller
variation in the CV responses for the training sub-sample, which in turn implies that a
brief training facilitates respondent ability to better process the risk information, and
13 However, the mean WTP (and not median WTP) for the sub-samples reduces between 1- 27 % with the
exclusion of these responses.
15
hence yields lower variation in the CV responses. We conduct more tests regarding the
differences in WTP in the econometric analysis in Section 4.
�
����,PSOLHG�YDOXH�RI�VWDWLVWLFDO�OLIH��We calculate the individual VSL from the data by dividing individual WTP by the
changes in the individually perceived risk; the mean VSL is in the US$ 1,783 to US$
2,922 range. As depicted in Table 6, we also calculate the individual VSL by dividing
the changes in the individual’ s average age-related mortality risk. We observe a
different pattern for the mean VSL (but not for the median VSL) based on the subjective
risk changes, between the trained and the not trained sub-samples. In the no-training
sub-sample, the mean VSL is higher for the higher (50%) risk reduction level, and in the
trained sub-sample, the mean VSL is higher for the lower (25%) risk reduction levels.
If the observed WTP were less than proportional to the magnitude of the risk change,
we would obtain a lower VSL for the higher risk reduction level. The result we
mentioned above is due to some very high WTPs related to very low perceived risks, in
the case of the no-training sub-sample. This pattern is not observed for VSL based on
objective risk changes; the mean VSL is lower for the higher risk reduction level.
The magnitude of the VSL, however, is very low compared to the available
estimates for developing countries.14 For example, using results from several VSL
studies, Miller (2000) predicts a VSL for Bangladesh in the range of US$ 30,000 to US$
1,000,000. The lower absolute values of VSL in our case may be attributed to the fact
that unlike many other studies we had relatively large risk reductions, if we assume that
14 Using data from the Indian labor market, Shanmugam (2000) provides VSL in the range of US $ 0.76
million -$1.026 million and Simon et al. (1999) provide VSL for India from an independent wage-risk
study in the range of US$ 0.15- US$ 0.35 million.
16
there is inadequate sensitivity to scope. For example, Carlsson et al. (2004) suggests that
VSL tends to decrease rapidly when the size of the risk reduction increases.
Moreover, stating WTP for a risk reduction is somewhat difficult for people
unfamiliar with the concept of trading income for risk reduction. Therefore, it is likely
that people would suffer from initial anchoring when constructing an answer as to how
much they would be willing to pay (Tversky and Kahneman, 1974; Kahneman et al.,
1982; Green and Tunstall, 2001). For example, the respondents might anchor on the
price of vaccination or on their other expenditures. People in rural areas, in a developing
country like Bangladesh, have a general perception that vaccinations are to be provided
free by the government, as is the case with the children’ s vaccination programs. For
example, Cropper et al. (2004) estimated demand for a hypothetical malaria vaccine in
rural Ethiopia and their results suggest that at very low prices few vaccines are
purchased; “ at an annualized price of US$ 3, half of the households in Tigary would
purchase no vaccines.” Moreover, in the context of a developing country, household
consumption expenditures are usually low on average, particularly in the rural areas.
Therefore, when placing a value on a desired and substantial risk reduction, the
respondents might anchor initially to such low expenditures, and adjust thereafter. In
addition, the incidence of financial limitations coupled with poorly functioning credit
markets also results in lower WTPs, particularly if the respondents are asked for one-
time payments rather than continuing monthly or yearly payments.15 Training seems to
reduce this potentially downward bias in WTP as we observed a significantly higher
WTP for the training sub-sample.
15 As observed by Carson (2000), “ A one-time payment generally produces more conservative estimates
since it does not offer the opportunity to spread payments over time,” compared to a continuing payment
(p. 1416).
17
���(FRQRPHWULF�DQDO\VLV�An appropriate econometric model for analyzing the WTP data that also includes zeros
would be Tobit with selection, as it allows for modeling zero and positive WTP
separately (for a discussion, see Carlsson and Johansson-Stenman, 2000). However,
with only 13 (non-protest) observations with a zero WTP, it is doubtful whether a
sample selection model can be justified in our case. Moreover, it is unclear whether
there is any true negative WTP. Therefore, we estimate a truncated regression model
where WTP is truncated at zero. Based on a PE test (p-value = 0.017) we can reject a
linear specification in favor of a log linear one. We use log (WTP + 1) as the dependent
variable, which is truncated at zero.
In the previous section, we observed that the distributions of WTP significantly
differ between the sub-samples concerning training. Given this result, we first estimated
a model allowing for heteroscedasticty concerning training, where dummy variables are
included identifying training in pooling the data. However, the heteroscedastic term is
not significantly different from zero (p value=0.96). Therefore, we cannot reject the null
hypothesis of homoscedasticity; the variances of WTP between the two sub-samples do
not differ significantly. Hence, we estimate the model assuming homoscedasticity and
dummy variables are included identifying training and risk reduction levels in pooling
the data. It is expected that people who passed the entire training would show higher
sensitivity to scope compared to other respondents; hence, we include separate
interaction variables.
We first estimate two models; one assuming that there are no enumerator effects on
WTP responses, and the other taking the enumerator effects into consideration by
including dummy variables for the enumerators (see e.g. Köhlin, 2001). Out of 13
18
enumerators, we found only one enumerator to be significant and negative. Based on a
likelihood ratio test, we can reject the null hypothesis of “ no enumerator effect” (p-
value 0.013). Therefore, we present the results from the latter model in Table 7, where
we control for the enumerator effects.
The coefficient “ training” is highly significant and implies that WTP for a larger
reduction in risk is 79% (e0. 58 – 1) higher for the group receiving the training. The
coefficient on a 50% reduction is highly significant and indicates that WTP for the
larger risk reduction is 45% higher than the WTP for the smaller reduction. The WTP
difference concerning risk reduction is even higher for the training sub-sample (16%
higher) in general and for the group who passed the tests (6% higher) in particular,
although these differences are not statistically significant. We find that people with
higher levels of prior education have on average 7% higher WTPs compared to illiterate
people; however, this difference is not statistically significant.
>>> TABLE 7 HERE
We also estimate a separate model controlling for the WTP difference
concerning the training as well as the risk reduction levels by interacting these variables
with respondents’ educational background variables. However, the interaction terms are
not significant, based on a likelihood ratio test (p-value 0.36); hence, we can accept the
results of the model presented in Table 7. Although it can be expected that people with
higher levels of education might have higher values for risk reduction, other studies, e.g.
Krupnick et al. (2002) and Alberini et al. (2004), have found that more highly educated
people report lower WTPs.
19
The estimated marginal effects for income16 are positive and significant, with an
income elasticity of 0.43. The result that the income elasticity of the VSL is well below
unity is also found in many other CV studies (e.g. Carlsson et al., 2004; Persson et al.,
2001). In cross-country comparisons of VSL studies, Miller (2000) as well as Viscusi
and Aldy (2003) found the income elasticity of VSL in the range of 0.85 to 1 and 0.5 to
0.6, respectively.
We find that the effect of age on WTP is negative at younger ages, until a
minimum is reached at age 46, and then increases. Alberini et al. (2004) found, using a
sample over 40-year olds that WTP does not decline until age 70. Although subjective
risk positively affects respondents’ WTPs for risk reduction, this is not significant.17
The result that subjective risk does not have any significant effect on WTP for risk
reduction is somewhat surprising, but consistent with other studies in developed
countries (e.g. Corso et al., 2001). We also observe that having a chronic illness has no
significant effect on WTP, which is consistent with the finding of Alberini et al. (2004)
that having a chronic condition does not reduce the WTP for mortality risk. Although
smokers’ risk perceptions were higher compared to non-smokers, being a smoker does
not significantly increase the WTP. We find that the level of overall individual
happiness significantly and positively affects WTP for risk reduction, all else remaining
the same.
16 It should be noted here that the distribution of income is highly skewed and hence we estimated
separate models excluding relatively high income. However, as the coefficient of income is roughly the
same, we decided to keep them in our final model presented here. 17 As respondent age and subjective risk might be correlated (risk perception is based on age-related
objective risk), we also estimate a separate model excluding the subjective risk. However, the results are
roughly the same.
20
From the estimated results, we formally test for the sensitivity to scope. We
calculate the mean WTP ratio, i.e. ratio of mean WTP for a 50% risk reduction to the
mean WTP for a 25% risk reduction, using the regression coefficients. The WTP ratios
for both the sub-samples are presented in Table 8. We obtain the standard errors for
these ratios as well as for the differences between these two ratios using the Delta
method (Greene, 2000) and hence, construct the corresponding confidence intervals.
�
>>> TABLE 8 HERE
We can reject the hypothesis that WTP is insensitive to the magnitude of risk
reduction for both the no-training and training sub-samples. We also find that sensitivity
to scope is higher (higher WTP ratio) for the trained sub-sample; however, the
difference between two WTP is not statistically significant.
�
�
�
���'LVFXVVLRQ�DQG�&RQFOXVLRQ�Past studies have discussed that it is problematic to consistently measure the value of
statistical life using the CV method, largely due to the cognitive burden that the
respondents face when comparing expected welfare effects of a small reduction in the
risk to those of small monetary changes (Beattie et al., 1998; Hammit and Graham,
1999; Carlsson et al., 2004). However, the main objective of this study was not to
obtain an absolute measure of VSL. Rather, we used substantial risk changes to be
21
valued and measured the effect of training regarding probability and risks on the WTP
responses and on the sensitivity to scope.
In the survey, we customize the mortality risk for respondents by asking them to
report their perception of the risk of dying based on their health and information about
age-related risks of dying provided in the survey. We find that people on average
overestimate the risk at younger ages and underestimate the risk at older ages. This
result is consistent with previous studies in the context of developed countries.
We find a significant difference in the WTP distributions between the sub-
sample receiving training and the sub-sample receiving no training. However, we find
no significant difference in the variance, but rather in the levels. We have found that
estimated WTP is sensitive to the size of the risk reduction in both the sub-samples.
Although sensitivity is higher for the trained sub-sample, the difference is not
statistically significant, which implies that the training does not affect the sensitivity to
scope. Although the implied VSL is higher for the trained sub-sample, it is still
substantially lower compared to other studies, which may be attributed to the fact that
compared to other studies we used relatively large risk reductions. Moreover, the
respondents in the survey were asked for a one-time payment, rather than for continuing
monthly or yearly payments, and there is possibility of respondents anchoring on the
often zero price that people pay for vaccination in reality.
Overall, it appears constructive to train the respondents regarding probabilities
and risk reductions. Training reduces the extent of cognitive burden that the respondents
face and thus increases the ability of the respondents to value the risk reduction in a
situation where the respondents are not familiar with the notion of probabilities and/or
risk/–money trade-offs. As discussed earlier, the respondents are likely to suffer from
22
initial anchoring in stating WTP for the risk reduction, which may be related to their
other expenditures that are usually low and which may result in a downward bias in
WTP. Training seems to reduce this potentially downward bias, since WTPs are
substantially higher in the trained sub-sample. However, there might also be some
problems with providing training in the CV survey. The respondents may get tired if
they find it boring and this may cause fatigue effects. Moreover, by talking a lot about
uncertainties and probabilities, the respondent can get the impression that avoiding risks
is very important. Hence, they will tend to state higher WTP in the training version.
This is then not because they are better trained but because they think that it is expected
of them. However, while some respondents may respond in this way, others are able to
draw inferences about the risk reduction, and training facilitates a cognitive structure
that is essential to draw such inference in such a situation.
Finally, using the CV method to elicit people’ s VSLs is not a “ mission
impossible.” CV risk-reduction can be performed in a developing country with very low
levels of education. A comprehensible training on probability and risk concepts,
interspersing risk examples with questions to maintain respondent interest as well as to
check understanding, should be given before presenting the CV scenario of risk
reduction to the respondents. There are remaining problems but most of these appear to
be related to the CV methodology SHU�VH, rather than to CV studies being performed in
developing countries.
�
5HIHUHQFHV�Alberini, A. (2004), Robustness of VSL Values from Contingent Valuation Surveys,
FEEM Working Paper No. 135.04, November.
23
Alberini, A., Cropper, M., Krupnick, A. and Simon, N. B. (2004), Does the value of a
statistical life vary with age and health status? Evidence from the US and
Canada, Journal of Environmental Economics and Management, 48(1): 769-792.
Balistreri, E., McClelland, G., Poe, G., and Schulze, W. (2001), Can Hypothetical
Questions Reveal True Values? A Laboratory Comparison of Dichotomous
Choice and Open-Ended Contingent Values with Auction Values,
Environmental and Resource Economics, 18: 275-292.
Bateman, I.J., Carson, R.T., Day, B., Hanemann, W.M., Hanely, N., Hett, T., Jones-Lee,
M., Loomes, G., Murato, S., Ozdemirogla, E., Pearce, D.W., Sugden, R. and
Swanson, J. (2002), Guidelines for the Use of Expressed Preferences Methods
for the Valuation of Preferences for Non-market Goods, Edward Elgar
Publishing, Cheltenham.
Beattie, J., Covey, P. Dolan, L. Hopkins, M, Jones-Lee, G., Pidgeon, A. Robinson, and
Spencer, A. (1998), On the Contingent Valuation of Safety and the Safety of
Contingent Valuation: Part 1- Caveat Investigator, Journal of Risk and
Uncertainty 17(3): 5-25
Carlsson, F. and Johansson-Stenman, O. (2000), Willingness to Pay for Improved Air
Quality in Sweden, Applied Economics, 32: 661-670.
Carlsson, F., Johansson-Stenman, O. and Martinsson, P. (2004), Is Transport Safety
More Valuable in the Air? , Journal of Risk and Uncertainty, 28:2; 147-163.
Carson, R.T. (2000), Contingent Valuation: A User’ s Guide, Environmental Science
and Technology, 34(8): 1413-1418.
Corso, P.S., Hammitt, J.K. and Graham, J. D. (2001), Valuing Mortality –Risk
Reduction: Using Visual Aids to Improve the Validity of Contingent Valuation,
Journal of Risk and Uncertainty, 23(2): 165-184.
Cropper, M.L., Halie, M., Lampietti, J., Poulos, C. and Whittington, D. (2004), The
demand for a malaria vaccine: evidence from Ethiopia, Journal of
Development Economics, 75:303-318
Diamond, P. A. and Hausman, J.A. (1994), Contingent Valuation: Is Some Number
Better Than No Number? The Journal of Economic Perspectives, 8(4), 45-64.
Gilovich, T., Griffin, D., and Kahneman, D (2002), Heuristics and Biases: The
Psychology of Intuitive Judgment, Cambridge, Cambridge University Press.
24
Green, C., and Tunstall, S. (2001), A Psychological Perspective, In: Bateman, I.J., and
Willis, K.G, (eds), Valuing Environmental Preferences, Theory and Practice of
the Contingent Valuation Method in the US, EU, and Developing Countries,
Oxford, Oxford University Press.
Greene, W. H. (2000), Econometric Analysis, Fourth Edition, Prentice Hall, USA.
Hammar, H. and Johansson-Stenman, O. (2004), The value of risk- free cigarettes- do
smokers underestimate the risk? Health Economics, 13: 59-71.
Hammitt, J.K. (2000), Valuing Mortality Risk: Theory and Practice, Environmental
Science and Technology, 34(8), 1396-1400.
Hammitt, J.K. and Graham, J. D. (1999) Willingness to pay for health protection:
Inadequate sensitivity to probability, Journal of Risk and Uncertainty 18, 33-62
Hanemann, W. M. (1994), Valuing the Environment through Contingent Valuation, The
Journal of Economic Perspectives, 8(4), 19-43.
Hanley, N., Ryan, M., and Wright, R. (2003), Estimating the monetary value of health
care: lessons from environmental economics, Health Economics, 12, 3-16.
ILO (2002), Cardiovascular System: Relative mortality risk for smokers,
http://www.ilo.org/public/protection/safework/tobacco/encyclo/car01fe.htm
Johansson, P.O. (2002), The value of a statistical life: theoretical and empirical
evidence, Applied Health Economics and Health Policy, 1(1), 33-41.
Kahneman, D. and Knetsch, J.L. (1992), Valuing Public Goods: The Purchase of Moral
Satisfaction, Journal of Environmental Economics and Management, 22(1), 57-
70.
Kahneman, D., and Tversky, A. (2000), Choices, Values and Frames, Cambridge,
Cambridge University Press.
Kahneman, D., Slovic, P. and Tversky, A. (1982), Judgment Under Uncertainty:
Heuristics and Biases, Cambridge, Cambridge University Press.
Köhlin, G. (2001), Contingent Valuation in project planning and evaluation: the case of
social forestry in Orissa, India, Environment and Development Economics,
6:237-258.
Krupnick, A., Alberini, A., Simon, N., O’ Brien, B., Goeree, R. and Heintzelman, M.
(2002), Age, Health and the Willingness to Pay for Mortality Risk Reductions:
25
A Contingent Valuation Survey of Ontario Residents, Journal of Risk and
Uncertainty, 24:2, 161-186.
Miller, T. R (2000), Variations between countries in values of statistical life, Journal of
Transport Economics and Policy, 34, 169-188.
Mitchell, R. C. and Carson, R.T. (1989), Using Surveys to Value Public Goods: The
Contingent Valuation Method, Washington D.C., Resource for the Future.
NOAA (1994), National Oceanic and Atmospheric Administration, Natural Resource
Damage Assessments: Proposed Rules, Federal Register 59, 1062-1191
Newcomb, P.A. and Carbone, PP (1992), The health consequences of smoking: cancer,
In: Fiore MC, Ed. Cigarette Smoking: A Clinical Guide to Assessment and
Treatment, Philadelphia, PA: WB Saunders Co: 305-331, Medical Clinics of
North America, http//www.cancer.org.
Persson, U., Norrinder, A., Hjalte, K. and Gralen, K. (2001), The Value of a Statistical
Life in Transport: Findings from a New Contingent Valuation Study in Sweden,
Journal of Risk and Uncertainty 23(2): 121-134.
Shanmugam, K.R. (2000), Valuations of Life and Injury Risks: Empirical Evidence
from India, Environmental and Resource Economics 16:379-389.
Shaw, M., Mitchell, R. and Dorling, D. (2000), Time for a smoke? One cigarette
reduces your life by 11 minutes, British Medical Journal, 320(1).
Siegel, S. and Castellan, N.J, 2000, Nonparametric statistics for the Behavioral
Sciences, McGraw-Hill.
Simon, N.B., Cropper, M. L., Alberini, A. and Arora, S. (1999), Valuing Mortality
Reductions in India: A Study of Compensating Wage Differentials, World Bank.
Slovic, P. (2000), The Perception of Risk, London, Earthscan.
Tversky, A. and Kahneman, D. (1974), Judgment Under Uncertainty: Heuristics and
Biases, Science, 185: 1124-1131
Tversky, A. and Kahneman, D. (1992), Advances in Prospect Theory: Cumulative
Representation of Uncertainty, Journal of Risk and Uncertainty, 5, 297-323
Viscusi, W. K., (1985), A Bayesian Perspective on Biases in Risk Perception,
Economics Letters 17, 59-62.
26
Viscusi, W.K., Magat, W. and Huber, J. (1987), An Investigation of the Rationality of
Consumer Valuations of Multiple Health Risk, RAND Journal of Economics,
18, 18, 465-479.
Viscusi, W. K. (1989), Prospective Reference Theory: Toward an Explanation of
Paradoxes, Journal of Risk and Uncertainty, 2, 235-263.
Viscusi, W. K. (1990), Do smokers underestimate risks? Journal of Political Economy,
98(6), 1253-1269
Viscusi, W. K. (1992), Smoking: Making the Risky Decision, Oxford University Press,
New York.
Viscusi, W.K. (1992), Fatal Tradeoffs: Public and Private Responsibilities for Risk,
New York, Oxford University Press.
Viscusi, W. K. (1993), The Value of Risks to Life and Health, Journal of Economic
Literature 31, 1912-1946.
Viscusi, W.K. and Aldy, J.E. (2003), The Value of a Statistical Life: A Critical Review
of Market Estimates throughout the World, Journal of Risk and Uncertainty,
27(1): 5-76.
Weinstein, M.C., Shepard, D.S., and Pliskin, J.S. (1980), The Economic Value of
Changing Mortality Probabilities: A Decision Theoretic Approach, Quarterly
Journal of Economics, 94(2), 373-396.
Whittington, D (1998), Administering Contingent Valuation Surveys in Developing
Countries, World Development, 26(1), 21-30.
Whittington, D. (2002), Improving the Performance of Contingent Valuation Studies in
Developing Countries, Environmental and Resource Economics, 22, 323-367.
WHO (2004), World Health Organization, Life Tables for 191 Countries, World
Mortality in 2000, http://www3.who.int/whosis/life_tables
27
Table 1. Sample Statistics (N=767)
Variable Mean Standard deviation
Minimum Maximum
Male 0.91 0.28 0 1
Muslim religion 0.66 0.47 0 1
Hindu religion 0.34 0.47 0 1
Age 43.6 12.4 19 75
Income per capita a 22,594 29,117 807 3,63,650
Illiterate (cannot read and write)
0.31 0.46 0 1
Low education (not illiterate and/or education up to high school level)
0.55 0.50 0 1
High education (education above high school level)
0.14 0.35 0 1
Having chronic illness b 0.39 0.49 0 1
Currently smoking 0.56 0.50 0 1
Self reported happiness c 5 2.2 0 10
a Total yearly household income was divided by [Number of adults + 0.5*number of children] 0.75, to
adjust for household size. N=765. Median 15508 Taka. 57.8 Taka=1 USD at the time of the survey
(October 2003).
b If the respondent has been suffering from any of the chronic diseases: heart disease, high blood
pressure, asthma, bronchitis, cancer, or diabetes.
c Responses, on an 11-point scale, to the question: “ As a whole, how happy would you say you are? The
scale is described as follows: 0 means “ extremely unhappy,” 10 means “ extremely happy’ ,” and 5
indicates average happiness such that half the population in Bangladesh is above 5 and half is below five.
28
Table 2. Understanding of probability and risk for the sub- sample with training
Probability/ risk questions a
% of respondents answered the test questions correctly
% of respondents answered correctly after 2nd explanation
% of respondents answered correctly after 3rd explanation
% of respondents never answered correctly
Dice throwing (PT1) 31 22 18 29
Lottery winning (PT2) 74 20 5 1
Mortality risk (PT3) 83 15 2 0.25
a See Figure 1 for exact wording of the test questions.
�
29
Table 3. Ordinary least square estimates of subjective risk
Dependent variable Log(subjective risk) Log(subjective risk)
Received training in the survey -0.104 (0.072)
-0.106 (0.072)
Log (age-related objective risk) 0.501*** (0.038)
Log (age) -0.548 (2.69)
Log(age)-squared 0.302 (0.362)
Having chronic illness 0.149** (0.074)
0.133* (0.075)
Low education -0.011 (0.083)
-0.018 (0.083)
High education -0.043 (0.123)
-0.050 (0.123)
Muslim religion 0.144* (0.076)
0.147* (0.076)
Log (income per capita) -0.070 (0.045)
-0.076 (0.046)
Currently smoking 0.185** (0.076)
0.153** (0.077)
Constant 2.08*** (0.462)
1.82** (5..01)
Adjusted R-square 0.206 0.220
Number of observations 765 765
Standard errors are in parentheses. Superscripts ***, **, * denote statistical significance at the 1 %, 5%,
and 10% level, respectively.
30
Table 4. Follow-up questions asked to respondents who stated zero WTP (N = 79)
Sub-sample of respondents who would want free vaccination (77%)
Sub-sample of respondents who would not want free vaccination (23%)
Reasons for not being willing to pay for vaccination a % of respondents agree
i) I cannot afford vaccinations, even though I believe it is good to have them.
79% -
ii) I think the government should pay for the vaccinations. 77% 11%
iii) I do not think the vaccine would really be safe. - 17%
iv) I do not think it is possible to reduce the mortality risk by vaccines.
7% 33%
v) I do not believe in reducing mortality risk by any means. 7% 44%
vi) Other reasons stated by the respondents: Reluctant to answer, not interested, dislike vaccination, not sure if (s) he would be willing to pay for vaccination.
3% 39%
a All the respondents who stated zero WTP were asked, “ Why would you not be willing to pay for
vaccinations?” . Then a list of possible reasons were read to them and the respondents were allowed
to choose more than one reason. They were also allowed to express other reasons.
31
Table 5. Probit regression of scenario rejection (N=79)
Variable Marginal effects Standard error
Received training in the survey -0.120 0.085
50% risk reduction -0.067 0.081
Age in years 0.005 0.002
Low education a 0.144 0.093
Muslim religion 0.416*** 0.145
Log (income per capita) 0.477 0.482
Having chronic illness -0.065 0.084
Currently smoking -0.138 0.109
Standard errors are in parentheses. Superscripts ***, **, * denote statistical significance at the 1 %, 5%,
and 10% level, respectively.
a We cannot estimate the marginal effect of high education as all eight observations from this group are
dropped because, for this group, all the zero WTPs imply a scenario rejection.
32
Table 6. WTP results for different sub samples a
No training Training
25% risk reduction
50% risk reduction
25% risk reduction
50% risk reduction
Sample size 162 168 175 189
Mean WTP 487 672 671 970
Standard deviation 1531 1934 1377 1324
Median WTP 100 200 500 500
Mean WTP ratio b 1.38 1.45
Null Hypothesis: Mean WTP ratio=1 c
p- value <0.001 p- value <0.001
96/�EDVHG�RQ�FKDQJHV�LQ��VXEMHFWLYH�ULVN
Mean VSL 1,03,074 1,06,585 1,68,905 1,07,697
Median VSL 20,000 13,333 33,333 30,000
95% confidence interval for mean VSL
43,742 – 1,62,407
32,164 – 1,81,005
1,03,714 – 1,34,097
81,167 - 1,34,228
96/�EDVHG�RQ�FKDQJHV�LQ��DYHUDJH�DJH�UHODWHG�REMHFWLYH�ULVN� Mean VSL 81,861 56,539 1,31,353 75,617
Median VSL 18,118 9,615 36,363 18,461
95% confidence interval for mean VSL
32,499 – 1,31,225
30,142 – 82,936 82,932 – 1,79,774 54,268 – 96,967
a WTP and VSL are expressed in Bangladesh Taka. 57.8 Taka =1 US $, at the time of survey (October
2003).
b Ratio of mean WTP for a 50% risk reduction to mean WTP for a 25% risk reduction.
c Using both the non-parametric Wilcoxon –Man-Whitney test and the W test.
33
Table 7. Estimated WTP by sub-samples: truncated regression model a
Dependent variable Log (WTP+1)
Variable Coefficient Standard error
Constant 15.12** 6.89
Received training 0.583*** 0.146
Passed probability test 0.170 0.230
50% risk reduction 0.371*** 0.137
Received Training × 50% risk reduction 0.146 0.203
Passed probability test × 50% risk reduction 0.060 0.310
Muslim religion -0.005 0.092
Log(age ) -7.02* 3.73
Log(age)-squared 0.912* 0.502
Log (subjective risk ) 0.021 0.052
High education 0.070 0.090
Low education -0.064 0.074
Log(income per capita) 0.425*** 0.063
Having chronic illness -0.010 0.102
Currently smoking 0.016 0.102
Self reported happiness 0.070*** 0.023
Disturbance standard deviation 1.22 0.033
Log-Likelihood -1100.12
Number of observations 692
Superscripts *** and ** denote statistical significance at the 1 % and 5% levels, respectively.
a We control for the enumerator effects but do not present the parameter estimates of the enumerator
dummies.
34
Table 8. Sensitivity to scope
No training Training
Mean WTP ratioa 1.45 1.51
Standard error 0.198 0.160
95 % confidence interval 1.06 - 1.84 1.20 – 1.83
Difference of mean WTP ratio 0.065
Standard error 0.073
95% confidence interval -0.08 – 0.21 a Ratio of mean WTP for a 50% risk reduction to mean WTP for a 25% risk reduction. The ratio is
calculated using the regression coefficients of the dummy variable for risk reduction and the mean
values of other explanatory variables.
35
Figure 1. Training - Probability and risk examples a
1RZ�,�ZLOO�GLVFXVV�WKH�FKDQFHV�DQG�ULVNV�RI�HYHQWV�RFFXUULQJ�XVLQJ�VRPH�H[DPSOHV��([DPSOH� ��� 6RPHWLPHV�ZH� WRVV� D� FRLQ� WR� GHFLGH� ZKLFK� RI� WZR� WKLQJV� WR� FKRRVH��:KHQ�ZH� WRVV� D� FRLQ�>(QXPHUDWRU��VKRZ�WRVVLQJ�D�FRLQ@��ZH�JHW�HLWKHU�D�KHDG�RU�D�WDLO��:H�FDQQRW�EH�VXUH�RI�WKH�UHVXOW�RI�WKH�WRVV��$V�WKHUH�DUH�WZR�WKLQJV�WKDW�FDQ�KDSSHQ�IURP�D�FRLQ�WRVV��WKH�FKDQFH�RI�JHWWLQJ�D�KHDG�LV���LQ����7KH�VDPH�LV�WUXH�IRU�JHWWLQJ�D�WDLO��6LPLODUO\��ZKHQ�ZH�UROO�D�GLFH��FKDNND��>(QXPHUDWRU��VKRZ�WKURZLQJ�D�FKDNND@�ZH�PD\�HLWKHU�VHH�RQ�WKH�WRS����������������RU����EXW�ZH�GRQ¶W� NQRZ�ZKLFK�RQH�EHIRUHKDQG��6LQFH� WKHUH�DUH�VL[�GLIIHUHQW� QXPEHUV�IURP���WR����ZH�PD\�VHH�DQ\�RI�WKHP�RQ�WKH�WRS��7KH�FKDQFH�RI�VHHLQJ�D���RQ�WKH�WRS�LV������,V�WKLV�H[DPSOH�FOHDU�WR�\RX"�>(QXPHUDWRU��,I�QR��H[SODLQ�DJDLQ�DQG�PDNH�VXUH�WKDW�WKH�UHVSRQGHQW�XQGHUVWDQGV��:ULWH�GRZQ�KRZ�PDQ\�WLPHV�\RX�KDG�WR�H[SODLQ��,I�WKH�UHVSRQGHQW�KDV�QRW�XQGHUVWRRG�DIWHU�WKUHH�WLPHV��ZULWH´��´�DQG�FRQWLQXH�@�37���1RZ��LI�,�WKURZ�WKLV�³FKDNND´��GLFH���ZKDW�LV�WKH�FKDQFH�WKDW���ZLOO�EH�VKRZQ�RQ�WRS"�$QVZHU��>(QXPHUDWRU��,I�WKH�DQVZHU�LV�ZURQJ��H[SODLQ�ZLWK�H[DPSOH�XQWLO�WKH�FRUUHFW�DQVZHU�LV�JLYHQ��:ULWH�GRZQ�KRZ�PDQ\� WLPHV� \RX� KDG� WR� H[SODLQ�� ,I� WKH� UHVSRQGHQW� GLG� QRW� KDYH� LW� ULJKW� DIWHU� D� WKLUG� H[SODQDWLRQ��H[SODLQ�WKH�DQVZHU�DQG�ZULWH����@�([DPSOH����&RQVLGHU�EX\LQJ�D�ORWWHU\�WLFNHW��0DQ\�SHRSOH�EX\�ORWWHU\�WLFNHWV�DQG�PRVW�SHRSOH�GR�QRW�ZLQ��6XSSRVH�WKDW�WKHUH�LV�RQO\���SUL]H�LQ�D�ORWWHU\�DQG�����SHRSOH�EX\�RQH�ORWWHU\�WLFNHW�HDFK��>(QXPHUDWRU��6KRZ�JULG�WDEOH��@���,Q�WKLV�FDVH�ZH�VD\�WKDW�WKH�FKDQFH�RI�ZLQQLQJ�WKH�SUL]H�ZLOO�EH���LQ������,V�WKLV�H[DPSOH�FOHDU�WR�\RX"�>(QXPHUDWRU��,I�QR��H[SODLQ�DJDLQ�DQG�PDNH�VXUH�WKDW�WKH�UHVSRQGHQW�XQGHUVWDQGV��:ULWH�GRZQ�KRZ�PDQ\�WLPHV�\RX�KDG�WR�H[SODLQ���,I�WKH�UHVSRQGHQW�KDV�QRW�XQGHUVWRRG�DIWHU�WKUHH�WLPHV��ZULWH´��´�DQG�FRQWLQXH�@�37���1RZ�� VXSSRVH� WKHUH� DUH� WZR� ORWWHULHV��7KH�FKDQFH� RI�ZLQQLQJ� LQ� RQH� ORWWHU\� LV� �� LQ� �����DQG� WKH�FKDQFH�RI�ZLQQLQJ�LQ�WKH�RWKHU�ORWWHU\�LV����LQ�������>(QXPHUDWRU��6KRZ�WKH�JULG�WDEOH����DQG�JULG�WDEOH����ZKHQ�H[SODLQLQJ@��:KLFK�ORWWHU\�KDV�WKH�ODUJHU�FKDQFH�RI�ZLQQLQJ"���LQ������ �����LQ������ ��$QVZHU��««««��>(QXPHUDWRU��,I�WKH�DQVZHU�LV�ZURQJ��H[SODLQ�ZLWK�H[DPSOH�XQWLO�WKH�FRUUHFW�DQVZHU�LV�JLYHQ��:ULWH�GRZQ�KRZ�PDQ\� WLPHV� \RX� KDG� WR� H[SODLQ�� ,I� WKH� UHVSRQGHQW� GLG� QRW� KDYH� LW� ULJKW� DIWHU� D� WKLUG� H[SODQDWLRQ��H[SODLQ�WKH�DQVZHU�DQG�ZULWH����@��([DPSOH����4XHVWLRQ�37���1RZ��VXSSRVH�WKHUH�DUH�WZR�URDGV�WKDW�DUH�ERWK�YHU\�SURQH�WR�DFFLGHQWV��7KH�ULVN�RI�G\LQJ�RQ�URDG�$�LV���LQ������DQG�WKH�ULVN�RI�G\LQJ�RQ�URDG�%�LV���LQ��������>(QXPHUDWRU��6KRZ�WKH�JULG�WDEOH���DQG�JULG�WDEOH����ZKHQ�H[SODLQLQJ@��:KLFK�URDG�LV�PRUH�ULVN\�WR�WDNH"�5RDG�$� � ��5RDG�%� � ��$QVZHU��«««�>(QXPHUDWRU��,I�WKH�DQVZHU�LV�ZURQJ��H[SODLQ�ZLWK�H[DPSOH�XQWLO�WKH�FRUUHFW�DQVZHU�LV�JLYHQ��:ULWH�GRZQ�KRZ�PDQ\� WLPHV� \RX� KDG� WR� H[SODLQ�� ,I� WKH� UHVSRQGHQW� GLG� QRW� KDYH� LW� ULJKW� DIWHU� D� WKLUG� H[SODQDWLRQ��H[SODLQ�WKH�DQVZHU�DQG�ZULWH����@�a The training is read by the enumerators.
36
Figure 2. Training - Explaining risk reduction
([DPSOH����6XSSRVH�WKH�DYHUDJH�ULVN�RI�G\LQJ�IRU�DQ�DGXOW�SHUVRQ�GXULQJ�WKH�QH[W���\HDUV�LV����LQ��������>(QXPHUDWRU��VKRZ�JULG�WDEOH���ZKHQ�H[SODLQLQJ@���6XSSRVH�D�UHGXFWLRQ�LQ�PRUWDOLW\�ULVN�� WKURXJK�VRPH�NLQG�RI�SXEOLF�PHDVXUH��FRXOG�UHGXFH�WKH�PRUWDOLW\�ULVN�IURP����LQ������WR����LQ������>(QXPHUDWRU��VKRZ�JULG�WDEOH���DQG�JULG�WDEOH���WRJHWKHU�WR�H[SODLQ�WKH�GLIIHUHQFH@���7KLV�PHDQV�WKDW��RQ�DYHUDJH����RXW�RI����ZRXOG�EH�VDYHG�E\�WKH�PHDVXUH��37���'R�\RX�XQGHUVWDQG�WKLV�ULVN�UHGXFWLRQ"�<HV� ��1R� ��$QVZHU���>(QXPHUDWRU��,I�QR��H[SODLQ�DJDLQ�DQG�PDNH�VXUH�WKDW�WKH�UHVSRQGHQW�XQGHUVWDQGV�DQG�ZULWH�GRZQ�KRZ�PDQ\�WLPHV�\RX�KDG�WR�H[SODLQ��,I�WKH�UHVSRQGHQW�KDV�QRW�XQGHUVWRRG�DIWHU�WKUHH�WLPHV��FRQWLQXH�DQG�ZULWH��@�([DPSOH�����6LPLODUO\��LI�WKH�ULVN�ZDV�UHGXFHG�IURP����LQ������WR����LQ�����>6KRZ�JULG�WDEOH���DQG�JULG�WDEOH���WRJHWKHU�WR�H[SODLQ�WKH�GLIIHUHQFH@��WKHQ����RXW�RI����ZRXOG�EH�VDYHG�RQ�DYHUDJH��37���'R�\RX�XQGHUVWDQG�WKLV�ULVN�UHGXFWLRQ"�<HV� ��1R� ��>(QXPHUDWRU��,I�QR��H[SODLQ�DJDLQ�DQG�PDNH�VXUH�WKDW�WKH�UHVSRQGHQW�XQGHUVWDQGV�DQG�ZULWH�GRZQ�KRZ�PDQ\�WLPHV�\RX�KDG�WR�H[SODLQ��,I�WKH�UHVSRQGHQW�KDV�QRW�XQGHUVWRRG�DIWHU�WKUHH�WLPHV��FRQWLQXH�DQG�ZULWH��@�$QVZHU�� � �([DPSOH���� ,I� WKH� ULVN�ZDV�UHGXFHG� IURP���� LQ������ WR���� LQ�����>VKRZ�JULG� WDEOH���DQG�JULG� WDEOH���WRJHWKHU�WR�H[SODLQ�WKH�GLIIHUHQFH@��WKHQ����RXW�RI����ZRXOG�EH�VDYHG�RQ�DYHUDJH��37���'R�\RX�XQGHUVWDQG�WKLV�ULVN�UHGXFWLRQ"�<HV� ��1R� ���>(QXPHUDWRU��,I�QR��H[SODLQ�DJDLQ�DQG�PDNH�VXUH�WKDW�WKH�UHVSRQGHQW�XQGHUVWDQGV�DQG�ZULWH�GRZQ�KRZ�PDQ\�WLPHV�\RX�KDG�WR�H[SODLQ��,I�WKH�UHVSRQGHQW�KDV�QRW�XQGHUVWRRG�DIWHU�WKUHH�WLPHV��FRQWLQXH�DQG�ZULWH��@�$QVZHU�� � � �([DPSOH���4XHVWLRQ�37����:KLFK�RI�WKH�DERYH�ULVN�UHGXFWLRQV�ZRXOG�\RX�SUHIHU"��>(QXPHUDWRU��6KRZ�WKH�FDUGV�DQG�OHW�WKH�UHVSRQGHQW�SRLQW@�D������LQ������WR����LQ������E������LQ������WR����LQ�������F������LQ������WR����LQ������
37
Figure 3. CV questionnaire: Risk perception
,W�KDV�EHHQ�HVWLPDWHG�WKDW�LQ�%DQJODGHVK��DQ�DYHUDJH�RI����RXW�RI������SHRSOH�LQ�WKH�������DJH�JURXS�ZLOO�GLH�RYHU�WKH�QH[W�ILYH�\HDUV�IURP�YDULRXV�FDXVHV��DQG����RXW�RI������SHRSOH�LQ�WKH�������DJH�JURXS�ZLOO�GLH�RYHU�WKH�QH[W�ILYH�\HDUV�IURP�YDULRXV�FDXVHV�(QXPHUDWRU��>VKRZ�JULG�WDEOH����DQG���@����5���7KLQNLQJ�DERXW�\RXU�RZQ�OLIH�DQG�WKH�ZD\�\RX�DUH�OLYLQJ�LW��ZKDW�GR�\RX�WKLQN�WKH�ULVN�RI�\RX�G\LQJ�LQ�WKH�QH[W�ILYH�\HDUV�LV"�>(QXPHUDWRU��/HW�WKH�UHVSRQGHQW�DOVR�VHH�WKH�WDEOHV����DQG����DJDLQ��DW�WKH�VDPH�WLPH@����$QVZHU��������������LQ������>(QXPHUDWRU��8VH�WKH�JULG�WDEOH�����ZKLFK�LV�DQ�HPSW\�JULG�WDEOH�WR�UHSUHVHQW�WKH�UHVSRQGHQW¶V�VXEMHFWLYH�ULVN�RI�G\LQJ�LQ�WKH�QH[W�ILYH�\HDUV��/HW�WKH�UHVSRQGHQW�ORRN�DW�LW�@�
38
Figure 4. Objective and subjective mortality risk during the next five years as a function of age
0
50
100
150
200
250
300
350
400
19-24 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64 65-69 70-74
Respondent age group (years)
Mor
talit
y ri
sk i
n 1
000 objective risk
subjective risk
39
Figure 5. CV scenario
3UHYHQWDWLYH�YDFFLQHV�FRXOG�UHGXFH�WKH�ULVN�RI�G\LQJ�IURP�PDQ\�LQIHFWLRXV�GLVHDVHV���6XSSRVH� WKDW� \RX� FRXOG� SDUWLFLSDWH� LQ� D� SURJUDP� LQYROYLQJ� YDULRXV� NLQGV� RI� YDFFLQDWLRQV� DJDLQVW�LQIHFWLRXV�GLVHDVHV��7KH�YDFFLQHV��LI�UHFHLYHG��ZRXOG�UHGXFH�\RXU�ULVN�RI�G\LQJ�GXULQJ�WKH�QH[W�ILYH�\HDUV����$VVXPH�WKDW�WKH�YDFFLQHV�ZRXOG�EH�FRPSOHWHO\�VDIH�DQG�ZRXOG�KDYH�QR�VLGH�HIIHFWV��+RZHYHU��WKH�HIIHFWV�RI�WKH�YDFFLQHV�ZRXOG�QRW�ODVW�EH\RQG�WKH�ILYH�\HDU�SHULRG���,I�UHFHLYHG��VXFK�YDFFLQHV�ZRXOG�UHGXFH�WKH�ULVN�RI�\RX�G\LQJ�RYHU�WKH�QH[W�ILYH�\HDUV�E\�RQH�TXDUWHU��RQH�KDOI��>(QXPHUDWRU��6KRZ�JULG�WDEOH����LQ�ZKLFK�WKH�VWDWHG�ULVN�IURP�TXHVWLRQ�&���KDV�EHHQ�LQFOXGHG�E\�ILOOLQJ�LQ�WKH�QXPEHU�RI�VTXDUHV�UHSUHVHQWLQJ� WKLV� VXEMHFWLYH�ULVN��6SOLW� WKH� ILOOHG� LQ�DUHD�LQWR����������������7KHQ� ZKLOH� PHQWLRQLQJ� WKH� ULVN� UHGXFWLRQ�� SRLQW� DW� WKH� ���� �� ���� SDUW� RI� WKH� VSOLW� ER[� DQG� ZKLOH�PHQWLRQLQJ�WKH�UHPDLQLQJ�ULVN�RQ�WKH�RWKHU�SDUW�RI�WKH�VSOLW�ER[�@��&9���:KDW�LV� WKH�PD[LPXP��DV�D�RQH�WLPH�IHH��\RX�ZRXOG�EH�ZLOOLQJ�WR�SD\�WR�REWDLQ�VXFK�YDFFLQHV�IRU�\RXUVHOI"�<RX�VKRXOG�DOVR�UHPHPEHU�WKDW�LI�\RX�ZHUH�WR�SD\�IRU�WKH�YDFFLQHV��\RX�ZRXOG�KDYH�OHVV�PRQH\�OHIW�IRU�RWKHU�SXUSRVHV��������������0D[LPXP�««««««7DND���
�
40
$SSHQGL[�����9LVXDO�DLG�IRU�ULVN�XQGHUVWDQGLQJ��
Figure 1. Grid table showing mortality risk of an adult in the next five years as 40 in 1000
41
$SSHQGL[�����'LVWULEXWLRQ�RI�:73��+LVWRJUDP��
Figure 1. Distribution of WTP for sub-sample: no training and 25% risk reduction
0.2
.4.6
.8Fr
actio
n
0 1000 2000 3000 4000 5000WTP
Figure 2. Distribution of WTP for sub-sample: training and 25% risk reduction
0.1
.2.3
.4Fr
actio
n
0 1000 2000 3000 4000 5000WTP
42
Figure 3. Distribution of WTP for sub-sample: no-training and 50% risk reduction
0.2
.4.6
Frac
tion
0 1000 2000 3000 4000 5000WTP
Figure 4. Distribution of WTP for sub-sample: training and 50% risk reduction
0.1
.2.3
.4Fr
actio
n
0 1000 2000 3000 4000 5000WTP
Recommended